Projects
Ongoing Projects
Team Sports Analytics (Cricket)
S Combinator Studio Pvt. Ltd. | Mar 2026 – Present
Working on developing computer vision models for cricket analytics, focusing on identifying phase transitions in gameplay and extracting key performance metrics for both batting and bowling. The goal is to provide interpretable insights that help players better understand their strategies and improve decision-making.
Major Completed Projects
Defect Analysis Model for Device Quality Inspection
Walmart & IIT Madras | Nov 2025 – Feb 2026
Working on a collaborative Walmart–IIT Madras project to develop a deep-learning–based defect analysis system that compares paired factory images to automatically identify subtle visual flaws and distinguish good devices from defective ones.
3D Human Motion Representation & Retrieval
TIME@ARC Hub & Griffith University | Oct 2025 – Dec 2025
Developed a scalable pipeline for structured 3D human motion representation with semantic alignment. Normalized large-scale skeletal data and applied clustering to extract representative motion patterns, enabling efficient retrieval and action recognition.s.
UMPIRE: Unsupervised Temporal Action Localization via Deep Clustering
IIT Gandhinagar | Jul 2025 – Oct 2025
Developed a fully label-free TAL framework that learns spatio-temporal graph embeddings from 3D skeleton sequences using ASTGCN and transformer-based temporal pooling, followed by DBSCAN clustering with adaptive ε-estimation.
- Achieved state-of-the-art performance on the BABEL dataset (51.53% mAP@IoU, 37.2 F1), surpassing prior unsupervised and weakly supervised methods.
Unsupervised Javelin Motion Phase Segmentation
SRIP Internship, IIT Gandhinagar | May 2025 – Jul 2025
Built an unsupervised ASTGCN + SOT framework to automatically detect biomechanical phase transitions in elite javelin throws, eliminating the need for manual labeling.
- Achieved state-of-the-art performance and released a dataset of 211 professionally annotated videos covering key motion phases.
Boxing Action Analysis
Research Project, IIT Gandhinagar | Feb 2025 – May 2025
Worked on a boxing action dataset, focusing on analyzing fine-grained motion patterns and temporal dynamics in combat sports. Explored techniques for action segmentation and performance understanding using computer vision methods.
Unsupervised Fine-Grained Action Localization in Sports Videos
Research Project, IIT Gandhinagar | Aug 2024 – Jan 2025
Designed an unsupervised skeleton-based pipeline using ASTGCN pre-training and an Action Dynamics Metric (ADM) to detect fine-grained motion boundaries in untrimmed sports videos.
- Achieved 82.67% mAP on the DSV Diving dataset — comparable to supervised methods — and demonstrated strong generalization to in-the-wild diving videos without labels.